AAAI 2025 Papers — Page 11
AAAI Conference on Artificial Intelligence · 3028 papers
Evaluating the Evaluator: Measuring LLMs’ Adherence to Task Evaluation Instructions
Bhuvanashree Murugadoss (Microsoft), Advait Sarkar (North Carolina State University)
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: This paper conducts an automatic evaluation of various large language models (GPT-4, Llama-3, Mistral, Phi-3, Prometheus-2) across eight public evaluation benchmarks by analyzing different levels of prompts (from no prompt, general prompts, specific indicator prompts to complete evaluation criteria prompts) and the perplexity of models without prompts. It proposes four major categories of evaluation criteria (content, relevance, completeness, engagement) to systematically analyze the consistency between model judgments and human annotations.
EvdCLIP: Improving Vision-Language Retrieval with Entity Visual Descriptions from Large Language Models
GuangHao Meng (Tsinghua University), Yong Jiang (Tsinghua University)
RetrievalTransformerLarge Language ModelVision Language ModelImageVideoTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: Utilize large language models to generate Entity Visual Descriptions (EVD) and inject them into queries to enhance semantic alignment in visual-language retrieval;
Eve: Efficient Multimodal Vision Language Models with Elastic Visual Experts
Miao Rang (Huawei Noah's Ark Lab), Yunhe Wang (Huawei Noah's Ark Lab)
RecognitionGenerationRetrievalComputational EfficiencyTransformerLarge Language ModelVision Language ModelImageTextMultimodality
🎯 What it does: An efficient visual language model, Eve, with approximately 1.8B parameters has been constructed, employing a three-stage training process and incorporating elastic visual experts, balancing both language and visual tasks.
Even-if Explanations: Formal Foundations, Priorities and Complexity
Gianvincenzo Alfano (University of Calabria), Irina Trubitsyna (University of Calabria)
Explainability and InterpretabilityGraph Neural Network
🎯 What it does: This paper studies the 'even-if' semifactual explanations from a theoretical perspective, providing formal definitions, complexity analysis, and proposing a user preference-based explanation priority framework that can generate semifactuals and factual explanations that align with user preferences.
Event-Enhanced Blurry Video Super-Resolution
Dachun Kai (University of Science and Technology of China), Xiaoyan Sun (University of Science and Technology of China)
RestorationSuper ResolutionConvolutional Neural NetworkOptical FlowVideo
🎯 What it does: This paper proposes an event camera-assisted blurry video super-resolution (BVSR) method called Ev-DeblurVSR, which can recover high-resolution, clear videos from low-resolution and blurry inputs.
Event2Tracking: Reconstructing Multi-Agent Soccer Trajectories Using Long-Term Multimodal Context
Harry Hughes (Stats Perform), Patrick Lucey (Queensland University of Technology)
Object TrackingTransformerMultimodalityTime Series
🎯 What it does: The Event2Tracking model is proposed, which reconstructs multi-agent football trajectories by integrating 60-second long broadcast tracking data with event data.
EventMamba: Enhancing Spatio-Temporal Locality with State Space Models for Event-Based Video Reconstruction
Chengjie Ge (University of Science and Technology of China), Zheng-Jun Zha (University of Science and Technology of China)
RestorationVideo
🎯 What it does: This paper proposes the EventMamba model, which improves the Mamba (SSM) module for the video reconstruction task of event cameras to maintain translational invariance and spatiotemporal locality, achieving efficient video reconstruction.
EventPillars: Pillar-based Efficient Representations for Event Data
Rui Fan (Xidian University), Zhangming Zhu (Xidian University)
RecognitionObject DetectionCompressionComputational EfficiencyConvolutional Neural NetworkTransformerImageTime Series
🎯 What it does: An efficient dense event representation framework called EventPillars based on pillars is proposed.
EventSum: A Large-Scale Event-Centric Summarization Dataset for Chinese Multi-News Documents
Mengna Zhu (National University of Defense Technology), Juanzi Li (Tsinghua University)
GenerationTransformerLarge Language ModelText
🎯 What it does: This paper proposes and implements the Event-Centric Multi-Document Summarization (ECS) task, which aims to automatically generate concise summaries that cover the core sub-events, time, location, people, and causal relationships of multiple related news articles.
EventZoom: A Progressive Approach to Event-Based Data Augmentation for Enhanced Neuromorphic Vision
Yiting Dong (University of Chinese Academy of Sciences), Yi Zeng (University of Chinese Academy of Sciences)
ClassificationRecognitionSpiking Neural NetworkContrastive LearningImageVideo
🎯 What it does: EventZoom is proposed, a progressive data augmentation method for event cameras aimed at enhancing the robustness of neuromorphic vision models.
Every Bit Helps: Achieving the Optimal Distortion with a Few Queries
Soroush Ebadian (University of Toronto), Nisarg Shah (University of Toronto)
🎯 What it does: This paper proposes a matching and election algorithm that can still achieve optimal social welfare approximation when each agent only queries a limited number of numerical queries (λ times).
Every Component Counts: Rethinking the Measure of Success for Medical Semantic Segmentation in Multi-Instance Segmentation Tasks
Alexander Jaus (Karlsruhe Institute of Technology), Rainer Stiefelhagen (Karlsruhe Institute of Technology)
SegmentationConvolutional Neural NetworkImageBiomedical DataComputed TomographyPositron Emission Tomography
🎯 What it does: An evaluation protocol based on connected components (CC-Metrics) is proposed, which calculates traditional semantic segmentation metrics (Dice, Hausdorff, Surface Dice, etc.) locally for each lesion, eliminating bias towards lesion size.
Everywhere Attack: Attacking Locally and Globally to Boost Targeted Transferability
Hui Zeng (Southwest University of Science and Technology), Anjie Peng (Southwest University of Science and Technology)
Adversarial AttackImage
🎯 What it does: Proposes and implements the 'everywhere attack', which significantly enhances the cross-model transferability of targeted attacks by simultaneously optimizing attack targets in both global and multiple local regions of the image.
EvHDR-GS: Event-guided HDR Video Reconstruction with 3D Gaussian Splatting
Zehao Chen (Zhejiang University), Gang Pan (Zhejiang University)
RestorationGaussian SplattingSimultaneous Localization and MappingVideo
🎯 What it does: Using event cameras and single-exposure LDR videos to construct HDR 3D Gaussian scenes, achieving full video-level HDR reconstruction, ensuring temporal consistency and eliminating data bias.
EvHDR-NeRF: Building High Dynamic Range Radiance Fields with Single Exposure Images and Events
Zehao Chen (Zhejiang University), Gang Pan (Zhejiang University)
RestorationGenerationNeural Radiance FieldImageVideo
🎯 What it does: Construct a high dynamic range (HDR) radiance field using single-exposure low dynamic range (LDR) images and event streams, capable of generating HDR or corresponding LDR views from arbitrary angles and exposure times.
EvoChart: A Benchmark and a Self-Training Approach Towards Real-World Chart Understanding
Muye Huang (Xi'an Jiaotong University), Jun Liu (Xi'an Jiaotong University)
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelTextMultimodalityBenchmark
🎯 What it does: Developed the EvoChart method and the EvoChart-QA benchmark, generating high-quality synthetic chart data through multi-stage self-training and training chart understanding models;
Evolutionary Classifier Chain for Multi-Dimensional Classification
Yu-Yang Zhang (Southeast University), Min-Ling Zhang (Lanzhou University of Technology)
ClassificationOptimizationTabular
🎯 What it does: A multi-dimensional classification method based on evolutionary algorithms, ECCO, is proposed, which can simultaneously optimize the order of the classifier chain and the feature subsets for each dimension, thereby improving the performance of multi-dimensional label prediction.
Evolutionary Large Language Model for Automated Feature Transformation
Nanxu Gong (Arizona State University), Yanjie Fu (Arizona State University)
OptimizationTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTabular
🎯 What it does: A novel evolutionary large language model framework (ELLM-FT) is proposed for automatic feature transformation, combining reinforcement learning to generate various population data and a small number of examples from LLM to create better feature combinations.
Evolutionary Reinforcement Learning with Parameterized Action Primitives for Diverse Manipulation Tasks
Xianxu Qiu (Shenzhen University), Fuchun Sun (Tsinghua University)
Robotic IntelligenceReinforcement Learning
🎯 What it does: This paper proposes a framework called ERLAP that combines evolutionary reinforcement learning with hierarchical reinforcement learning, utilizing parameterized action primitives to generate sequences and improving exploration and mitigating catastrophic forgetting through a dual-population evolutionary mechanism.
EvSTVSR: Event Guided Space-Time Video Super-Resolution
Haojie Yan (Zhejiang University), Gang Pan (Zhejiang University)
RestorationSuper ResolutionTransformerOptical FlowVideoMultimodality
🎯 What it does: This paper proposes the EvSTVSR method, which utilizes the high temporal resolution of event cameras and a small number of RGB frames to achieve spatial-temporal video super-resolution.
EWMoE: An Effective Model for Global Weather Forecasting with Mixture-of-Experts
Lihao Gan (University of Electronic Science and Technology of China), Jie Shao (University of Electronic Science and Technology of China)
TransformerMixture of ExpertsTime Series
🎯 What it does: The EWMoE model is proposed for global weather forecasting, achieving high-accuracy predictions using only two years of ERA5 data while significantly reducing training resources.
Exact Algorithms and Lower Bounds for Forming Coalitions of Constrained Maximum Size
Foivos Fioravantes (Czech Technical University in Prague), Nikolaos Melissinos (Czech Technical University in Prague)
OptimizationGraph
🎯 What it does: This paper studies the optimal coalition partition problem of additive separable hedonic games (ASHG) under a given maximum team size constraint, analyzes its parameterized complexity, and provides various FPT algorithms and kernelization results.
Exact Algorithms for Multiagent Path Finding with Communication Constraints on Tree-Like Structures
Foivos Fioravantes (Czech Technical University in Prague), Michal Opler (Czech Technical University in Prague)
🎯 What it does: The study investigates the parameterized complexity of the multi-agent pathfinding problem under communication constraints, proposing FPT algorithms for graph structures such as trees, tree-width, and bounded degree, and proving W[1]-hardness when only the number of agents is given.
EXCGEC: A Benchmark for Edit-Wise Explainable Chinese Grammatical Error Correction
Jingheng Ye (Tsinghua University), Wenhao Jiang (Sun Yat-Sen University)
Explainability and InterpretabilityTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: The EXGEC task is proposed, and the EXCGEC Chinese interpretable grammar error correction benchmark is constructed, providing editing-style explanations and training and evaluation of multi-task models.
Excluding the Impossible for Open Vocabulary Semantic Segmentation
Shiyuan Zhao (China University of Petroleum), Shuai Shao (China University of Petroleum)
SegmentationContrastive LearningImage
🎯 What it does: Using a reverse approach of 'excluding the impossible', combined with CLIP/CLIPN, ELSE-Net is constructed to achieve open vocabulary semantic segmentation.
ExcluIR: Exclusionary Neural Information Retrieval
Wenhao Zhang (Shandong University), Pengjie Ren (Shandong University)
RetrievalTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: Constructed the ExcluIR dataset and benchmark to evaluate the performance of retrieval models on exclusive queries, and conducted experiments on various retrieval models.
ExDDI: Explaining Drug-Drug Interaction Predictions with Natural Language
Zhaoyue Sun (University of Warwick), Yulan He (King's College London)
Explainability and InterpretabilityDrug DiscoveryTransformerLarge Language ModelSupervised Fine-TuningTextBiomedical DataRetrieval-Augmented Generation
🎯 What it does: This paper proposes a natural language explanation task for generating drug-drug interaction (DDI) predictions and constructs a series of ExDDI models that can generate explanations for both positive and negative interactions while making predictions.
Expand Horizon: Graph Out-of-Distribution Generalization via Multi-Level Environment Inference
Jiaqiang Zhang (Nanjing University of Aeronautics and Astronautics), Songcan Chen (Nanjing University of Aeronautics and Astronautics)
Domain AdaptationGraph Neural NetworkTransformerGraph
🎯 What it does: In the scenario of node-level distribution shift, a multi-layer environment inference model (MLEI) is proposed to capture multi-scale environmental information of graph structures and enhance out-of-bag (OOB) generalization performance.
Expand VSR Benchmark for VLLM to Expertize in Spatial Rules
Peijin Xie (Harbin Institute of Technology), Jiajia Zhang (Harbin Institute of Technology)
GenerationData SynthesisSuper ResolutionTransformerVision Language ModelDiffusion modelImageMultimodalityBenchmark
🎯 What it does: This paper constructs the visual large language model VSR expert VSRE by extending VSR data and visual encoding.
Expanding the Scope of Negatives: Boosting Image-Text Matching with Negatives Distribution Guided Learning
Zhao Zhou (Fudan University), Cheng Jin (Fudan University)
RetrievalContrastive LearningImageText
🎯 What it does: A negative sample distribution guided image-text matching framework (NDGL) is proposed, which utilizes the semantic distribution information of all negative samples to assist training while maintaining focus on the hardest negative samples.
Expected Hypervolume Improvement Is a Particular Hypervolume Improvement
Jingda Deng (Xi'an Jiaotong University), Hui Li (Xi'an Jiaotong University)
OptimizationGaussian SplattingPoint Cloud
🎯 What it does: This paper proposes to rewrite the Expected Hypervolume Improvement (EHVI) and its batch version (q EHVI) into a specific form of hypervolume improvement, providing the corresponding analytical expressions.
Expensive Multi-Objective Bayesian Optimization Based on Diffusion Models
Bingdong Li (East China Normal University), Aimin Zhou (East China Normal University)
OptimizationDiffusion model
🎯 What it does: A Pareto set learning algorithm based on a combined diffusion model, CDM-PSL, is proposed to address expensive multi-objective Bayesian optimization problems.
ExPERT: Modeling Human Behavior Under External Stimuli Aware Personalized MTPP
Subhendu Khatuya (Indian Institute of Technology Kharagpur), Niloy Ganguly (Indian Statistical Institute)
TransformerTime SeriesSequential
🎯 What it does: This study investigates the introduction of external stimuli into continuous-time event sequences and the implementation of personalized modeling through a marked point process framework.
Explainable Neural Networks with Guarantee: A Sparse Estimation Approach
Antoine Ledent (Singapore Management University), Peng Liu (Singapore Management University)
Explainability and InterpretabilityTabular
🎯 What it does: A sparse interpretable neural network called SparXnet is designed, achieving joint learning of feature selection and one-dimensional transformation, balancing interpretability and predictive performance.
Explaining Decisions of Agents in Mixed-Motive Games
Maayan Orner (Bar-Ilan University), Sarit Kraus (Bar-Ilan University)
Explainability and InterpretabilityLarge Language ModelTabular
🎯 What it does: This study proposes three explanation methods (SBUE, SICA, and possible action explanation) to elucidate the decision-making of agents in multi-agent mixed-motive games.
Explanation Bottleneck Models
Shin'ya Yamaguchi (NTT), Kosuke Nishida (NTT)
ClassificationExplainability and InterpretabilityKnowledge DistillationTransformerVision Language ModelImageMultimodality
🎯 What it does: An explainable bottleneck model (XBM) is proposed, which directly generates natural language explanations through a pre-trained vision-language encoder-decoder to predict final labels, eliminating the dependence on a predefined set of concepts.
Explicit and Implicit Examinee-Question Relation Exploiting for Efficient Computerized Adaptive Testing
Changqian Wang (Anhui University), Bo Jin (Dalian University of Technology)
OptimizationReinforcement LearningGenerative Adversarial NetworkContrastive LearningTabular
🎯 What it does: An adaptive testing framework RECAT is proposed, which integrates explicit and implicit examinee-item relationships, utilizing a generative item selector and candidate set matching to improve item selection efficiency and accuracy.
Explicit and Implicit Graduated Optimization in Deep Neural Networks
Naoki Sato (Meiji University), Hideaki Iiduka (Meiji University)
OptimizationImage
🎯 What it does: Evaluated the performance of explicit gradient optimization algorithms on traditional benchmark functions and deep neural networks, and proposed an implicit gradient optimization algorithm using SGD with momentum, providing theoretical convergence analysis and experimental validation.
Explicit Relational Reasoning Network for Scene Text Detection
Yuchen Su (Fudan University), Xieping Gao (Hunan Normal University)
RecognitionObject DetectionTransformerImage
🎯 What it does: This paper proposes ERRNet, treating scene text detection as a tracking problem, and implements end-to-end text detection without post-processing using an explicit relationship reasoning network.
Explicitly Guided Difficulty-Controllable Visual Question Generation
Jiayuan Xie (Hong Kong Polytechnic University), Qing Li (Chongqing University)
GenerationTransformerLarge Language ModelImageText
🎯 What it does: Generate visual questions of varying difficulty based on object relationship chains in images, proposing a controllable multi-step question generation framework called MultiStepGen.
Exploit Gradient Skewness to Circumvent Byzantine Defenses for Federated Learning
Yuchen Liu (Zhejiang University), Gang Chen (Zhejiang University)
Federated LearningAdversarial AttackImage
🎯 What it does: The study investigates the phenomenon of gradient skewness in non-IID federated learning environments and proposes a two-stage attack method based on gradient skewness called STRIKE, aimed at bypassing existing Byzantine defenses.
Exploit Your Latents: Coarse-Grained Protein Backmapping with Latent Diffusion Models
Rongchao Zhang (Peking University), Hanpin Wang (Chinese PLA General Hospital)
Protein Structure PredictionGraph Neural NetworkTransformerDiffusion modelContrastive LearningBiomedical Data
🎯 What it does: This study investigates a coarse-grained protein remapping method based on a latent diffusion model, proposing LatCPB, which achieves high-resolution remapping from coarse-grained to atomic level through discrete latent space, contrastive learning, and conditional latent diffusion.
Exploiting Continuous Motion Clues for Vision-Based Occupancy Prediction
Haoran Xu (Shenzhen Campus of Sun Yat-sen University), Luntong Li (Shenzhen Campus of Sun Yat-sen University)
Object DetectionObject TrackingAutonomous DrivingConvolutional Neural NetworkRecurrent Neural NetworkOptical FlowImagePoint Cloud
🎯 What it does: A continuous updating occupancy prediction framework (CMOP) is proposed, which dynamically iterates and updates the 3D occupancy volume by combining historical occupancy information with real-time optical flow information.
Exploiting Diffusion Prior for Real-World Image Dehazing with Unpaired Training
Yunwei Lan (University of Science and Technology of China), Dong Liu (University of Science and Technology of China)
Image TranslationRestorationDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes Diff-Dehazer, a dehazing framework based on unpaired training, which uses the pre-trained Stable Diffusion as a bidirectional mapping network of CycleGAN, achieving more natural dehazing through physical and textual priors.
Exploiting Fine-Grained Skip Behaviors for Micro-Video Recommendation
Sanghyuck Lee (Chung Ang University), Jaesung Lee (Chung Ang University)
Recommendation SystemGraph Neural NetworkReinforcement LearningVideo
🎯 What it does: This study investigates a recommendation model that utilizes fine-grained classification of skip behaviors in micro-videos (highly positive, mildly positive, negative) to construct a dual-graph structure, aiming to enhance the capture of user viewing preferences.
Exploiting Multimodal Spatial-temporal Patterns for Video Object Tracking
Xiantao Hu (Nanjing University), Jian Yang (Nanjing University)
Object TrackingTransformerVideoMultimodality
🎯 What it does: A unified tracking framework STTrack based on multimodal spatiotemporal patterns is proposed, capable of continuously capturing target motion information and achieving precise localization across various modalities such as RGB, TIR, Depth, and Event.
Exploiting Symmetries in MUS Computation
Ignace Bleukx (KU Leuven), Tias Guns (KU Leuven)
OptimizationTabularBenchmark
🎯 What it does: This paper studies and implements a method to accelerate the computation of minimal unsatisfiable subsets (MUS) using symmetry.
Explore In-Context Segmentation via Latent Diffusion Models
Chaoyang Wang (Peking University), Shuicheng Yan (Nanyang Technological University)
SegmentationDiffusion modelImageVideoBenchmark
🎯 What it does: This paper proposes an unsupervised visual context learning segmentation framework based on the Latent Diffusion Model (LDM), called LDIS, which achieves in-context segmentation of images and videos using only visual prompts instead of text prompts or additional networks.
Explore What LLM Does Not Know in Complex Question Answering
Xin Lin (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
RetrievalOptimizationTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: A RAG framework KEQA based on question-answer knowledge assessment is proposed, which can accurately identify knowledge gaps in LLMs and only retrieve missing knowledge, thereby improving reasoning quality.
Exploring Activation Patterns of Parameters in Language Models
Yudong Wang (Peking University), Zhifang Sui (Peking University)
Large Language ModelText
🎯 What it does: By evaluating parameter activation based on gradient first-order metrics, we analyze the activation distribution of different layers under same-domain and cross-domain inputs, and propose the LLMDcos metric.
Exploring Conversational Adaptability: Assessing the Proficiency of Large Language Models in Dynamic Alignment with Updated User Intent
Yu-Chuan Chen (Academia Sinica), Hen-Hsen Huang (Academia Sinica)
TransformerLarge Language ModelText
🎯 What it does: This paper studies the ability of large language models (LLMs) to capture and update user intentions in real-time during conversations, and based on this, proposes an evaluation framework that does not use Chain-of-Thought (CoT) or prompt engineering.
Exploring Enhanced Contextual Information for Video-Level Object Tracking
Ben Kang (Dalian University of Technology), Dong Wang (Baidu Inc.)
Object TrackingVideo
🎯 What it does: This paper proposes the MCITrack framework, which utilizes Mamba's hidden states to continuously record and transmit video-level contextual information to enhance the accuracy of visual object tracking.
Exploring Model Editing for LLM-based Aspect-Based Sentiment Classification
Shichen Li (Soochow University), Peifeng Li (Westlake University)
ClassificationDomain AdaptationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a framework that combines causal tracking and model editing, achieving domain adaptation for sentiment polarity detection with minimal parameter updates through targeted editing of weights and representation layers in large language models.
Exploring More from Multiple Gait Modalities for Human Identification
Dongyang Jin (Southern University of Science and Technology), Shiqi Yu (Alibaba Group)
RecognitionOptical FlowVideoMultimodality
🎯 What it does: This paper studies and compares the roles of three pedestrian gait modalities (shape, human segmentation, optical flow) in multimodal gait recognition and proposes a C2 Fusion method based on shared and differential features.
Exploring Query Efficient Data Generation Towards Data-Free Model Stealing in Hard Label Setting
Gaozheng Pei (University of Chinese Academy of Sciences), Yingfei Sun (University of Chinese Academy of Sciences)
GenerationData SynthesisKnowledge DistillationAdversarial AttackConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper addresses the issue of model stealing without data under hard label settings, proposing a Query Efficient Data Generation (QEDG) method that enhances stealing efficiency by generating samples that are close to and evenly distributed around the decision boundary of the target model.
Exploring Rationale Learning for Continual Graph Learning
Lei Song (Southeast University), Youyong Kong (Southeast University)
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: A graph-level continual learning framework RL-GNN is proposed, which introduces the principle of invariant learning into graph neural networks to extract the core 'rationale' of graphs, thereby alleviating catastrophic forgetting.
Exploring Salient Object Detection with Adder Neural Networks
Bo-Wen Yin (Nankai University), Zheng Lin (Tsinghua University)
Object DetectionComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A significant object detection model is developed using a lower energy-consuming Additive Neural Network (ANN), and the issue of insufficient feature diversity of ANN in the decoder is analyzed through experiments. A Differential Merging Module (DMM) and Differential Channel Attention (DCA) are proposed to enhance performance.
Exploring Semantic Consistency and Style Diversity for Domain Generalized Semantic Segmentation
Hongwei Niu (Xiamen University), Shengchuan Zhang (Xiamen University)
SegmentationDomain AdaptationTransformerContrastive LearningImage
🎯 What it does: This paper proposes a framework named SCSD for domain generalization in semantic segmentation, which combines three main modules: semantic consistency enhancement, text-driven style transformation, and style collaborative optimization, significantly improving the model's generalization ability to unknown target domains.
Exploring Task-Level Optimal Prompts for Visual In-Context Learning
Yan Zhu (Tianjin University), Changqing Zhang (Tianjin University)
Object DetectionSegmentationPrompt EngineeringImage
🎯 What it does: This study investigates task-level prompt selection in visual context learning and proposes two untrained prompt search strategies.
Exploring Temporal Event Cues for Dense Video Captioning in Cyclic Co-Learning
Zhuyang Xie (Southwest Jiaotong University), Xiao Wu (Southwest Jiaotong University)
GenerationRetrievalTransformerContrastive LearningVideoTextMultimodality
🎯 What it does: A multi-concept cyclic learning (MCCL) framework is proposed for dense video description, combining video-text retrieval, frame-level weakly supervised concept detection, and a cyclic co-learning mechanism between the generator and the locator.
Exploring the Better Multimodal Synergy Strategy for Vision-Language Models
Xiaotian Yin (University of Science and Technology of China), Tianzhu Zhang (University of Science and Technology of China)
ClassificationDomain AdaptationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: A Dual-shared Adapter (DsRA) based on LoRA is proposed for cross-modal collaboration in few-shot learning with CLIP.
Exploring the Potential of Large Vision-Language Models for Unsupervised Text-Based Person Retrieval
Zongyi Li (Huazhong University of Science and Technology), Hefei Ling (Huazhong University of Science and Technology)
RetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Using various large-scale visual-language models (LVLM) to generate multi-granularity and multi-style pseudo-text descriptions, and training a text-image retrieval model in an unsupervised setting with only image data through an uncertainty modeling and alignment framework (MUMA).
Exploring the Relationship Between Samples and Masks for Robust Defect Localization
Jiang Lin (Southeast University), Yaping Yan (Southeast University)
Anomaly DetectionGenerative Adversarial NetworkImage
🎯 What it does: Through an unsupervised GAN framework, utilizing visible pattern loss and a dynamic correction mechanism, directly locate texture defects without modeling normal distribution and defect labeling.
Exploring Unbiased Deepfake Detection via Token-Level Shuffling and Mixing
Xinghe Fu (Zhejiang University), Xi Li (Tencent)
ClassificationRecognitionTransformerContrastive LearningVideo
🎯 What it does: This paper proposes an unbiased deepfake detection framework based on Vision Transformer, which disrupts and mixes position and content biases at the token level, and utilizes alignment loss to achieve unbiased feature and classifier learning.
Exploring Vacant Classes in Label-Skewed Federated Learning
Kuangpu Guo (University of Science and Technology of China), Tieniu Tan (Nanjing University)
Federated LearningKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: In federated learning, to address the issue of missing categories and the decline in recognition rates for minority categories caused by uneven label distribution, the FedVLS method is proposed, which simultaneously applies knowledge distillation for missing categories and logits suppression during local training.
Exponential-Family Harmoniums with Neural Sufficient Statistics
Azwar Abdulsalam (Purdue University), Joseph G. Makin (Purdue University)
GenerationData SynthesisConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes a neural network-based sufficient statistic exponential family harmonization model (NN‑EFH) and trains and generates on complex image datasets using a Langevin-within-Gibbs sampling scheme.
Expressive Power of Temporal Message Passing
Przemysław Andrzej Wałęga (Queen Mary University of London), Michael Rawson (University of Southampton)
Graph Neural NetworkGraphTime Series
🎯 What it does: This paper classifies and analyzes the expressive power of graph neural networks for temporal message propagation mechanisms.
External Reliable Information-enhanced Multimodal Contrastive Learning for Fake News Detection
Biwei Cao (Southeast University), Jie Gui (Southeast University)
ClassificationRepresentation LearningConvolutional Neural NetworkRecurrent Neural NetworkContrastive LearningTextMultimodality
🎯 What it does: This paper proposes an external reliable information-enhanced multimodal contrastive learning framework ERIC-FND, which enriches text representation with entity descriptions and improves fake news detection through multimodal contrast and cross-modal interaction.
Extract Free Dense Misalignment from CLIP
JeongYeon Nam (NAVER Cloud AI), Taeho Kil (NAVER AI Lab)
RetrievalExplainability and InterpretabilityTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: Using pre-trained CLIP to detect fine-grained mismatches (such as word-level mismatches) between images and text through gradient attribution methods.
Extracting Interpretable Task-Specific Circuits from Large Language Models for Faster Inference
Jorge García-Carrasco (University of Alicante), Juan Trujillo (University of Alicante)
Explainability and InterpretabilityComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelText
🎯 What it does: This paper proposes an automated pruning method based on Mechanism Interpretability (MI), which can extract the minimal sub-circuit from large language models (LLMs) that is only used for executing specific tasks, enabling independent inference of the sub-model without additional training or fine-tuning.
Extracting PAC Decision Trees from Black Box Binary Classifiers: The Gender Bias Study Case on BERT-based Language Models
Ana Ozaki (University of Oslo), Anders Imenes (University of Bergen)
ClassificationExplainability and InterpretabilityTransformerLarge Language ModelText
🎯 What it does: This paper studies a decision tree extraction method based on PAC theory to approximate black-box binary classifiers and detect occupational gender bias in BERT-like language models.
EyEar: Learning Audio Synchronized Human Gaze Trajectory Based on Physics-Informed Dynamics
Xiaochuan Liu (Renmin University of China), Denghao Zhang (Renmin University of China)
Recurrent Neural NetworkContrastive LearningImageMultimodalityPhysics RelatedAudio
🎯 What it does: Designed an audio-synchronized viewing task and constructed a dataset of 20,000 fixation points from 8 participants, proposing the EyEar framework based on physical information dynamics to predict human eye gaze trajectories in visual scenes.
FaceA-Net: Facial Attribute-Driven ID Preserving Image Generation Network
Jiayu Wang (Fudan University), Yu-Gang Jiang (Microsoft Research)
GenerationData SynthesisSupervised Fine-TuningDiffusion modelImageText
🎯 What it does: By fine-tuning on a small number of images of the same person's identity, a high-fidelity image generation of facial identity is achieved using a facial attribute-driven training method and an ID-context decoupling framework, supporting fine-grained control over expressions, accessories, styles, and more.
FaceMe: Robust Blind Face Restoration with Personal Identification
Siyu Liu (Nankai University), Chongyi Li (Nankai University)
RestorationGenerationDiffusion modelImage
🎯 What it does: We propose FaceMe, a personalized blind face restoration method that does not require fine-tuning and supports an arbitrary number of reference images.
FaceSpeak: Expressive and High-Quality Speech Synthesis from Human Portraits of Different Styles
Tian-Hao Zhang (University of Science and Technology Beijing), Xu-Cheng Yin (University of Science and Technology Beijing)
GenerationData SynthesisContrastive LearningImageMultimodalityAudio
🎯 What it does: This paper proposes FaceSpeak, a multimodal TTS system based on multi-style portraits that decouples expression and identity, capable of generating corresponding speech based on images of different styles.
Facility Location Games with Optional Preferences: A Revisit
Xingchen Sha (City University of Hong Kong), Minming Li (City University of Hong Kong)
Optimization
🎯 What it does: This paper studies the optional preference location game for k (k≥3) facilities in a one-dimensional linear space and designs a strategyproof deterministic mechanism to satisfy agents' optional facility preferences while approaching the goal of minimizing agents' maximum/total costs.
Factor Augmented Tensor-on-Tensor Neural Networks
Guanhao Zhou (University of Notre Dame), Xiufan Yu (University of Notre Dame)
Time SeriesMagnetic Resonance ImagingAgriculture Related
🎯 What it does: This paper proposes a Tensor-on-Tensor prediction framework (FATTNN) that combines tensor factor models with deep neural networks for multi-step prediction of multi-dimensional tensor time series.
FactorGCL: A Hypergraph-Based Factor Model with Temporal Residual Contrastive Learning for Stock Returns Prediction
Yitong Duan (Tsinghua University), Jian Li (Tsinghua University)
Recommendation SystemOptimizationRecurrent Neural NetworkGraph Neural NetworkContrastive LearningTabularTime SeriesFinance Related
🎯 What it does: This paper proposes a hypergraph-based factor model, FactorGCL, for stock return prediction.
Fair and Efficient Completion of Indivisible Goods
Vishwa Prakash HV, Rohit Vaish (Indian Institute of Technology Delhi)
🎯 What it does: This paper proposes the completion problem of fairness and efficiency under 'frozen resources', studying whether the distribution of remaining items based on partially allocated goods (frozen goods) can meet the requirements of fairness (EF1, Prop1, MMS) and efficiency (Pareto optimality).
Fair Division via the Cake-Cutting Share
Yannan Bai (Duke University), Ian Zhang (Duke University)
Tabular
🎯 What it does: This paper proposes new concepts of fair shares based on individuals' utilities of other players, such as 'cake cutting shares' and 'envy-free shares', and studies the achievable approximation factors in the worst-case scenario.
Fair Division with Market Values
Siddharth Barman (Indian Institute of Science), Nisarg Shah (University of Toronto)
🎯 What it does: A model for fair distribution under the premise of market value is proposed, along with various existence and infeasibility results regarding fairness and efficiency.
Fair Division with Social Impact
Michele Flammini (Gran Sasso Science Institute), Giovanna Varricchio (University of Calabria)
Optimization
🎯 What it does: The paper studies how to maximize social utility (using the utility sum of social impact functions) while ensuring fairness (such as EF1, EFX, PROP1, epistemic EF1, etc.) in the context of resource allocation's impact on society.
Fair Federated Survival Analysis
Md Mahmudur Rahman (University of Maryland), Sanjay Purushotham (University of Maryland)
Federated LearningBiomedical Data
🎯 What it does: This paper proposes FairFSA, a fair federated survival analysis framework that combines FPV and DRO to train fair and debiased survival prediction models on multi-institutional data.
Fair Text-to-Image Diffusion via Fair Mapping
Jia Li (King Abdullah University of Science and Technology), Di Wang (Institute of Information Engineering, Chinese Academy of Sciences)
GenerationData SynthesisDiffusion modelImageText
🎯 What it does: This paper proposes Fair Mapping, a lightweight linear mapping network and detector inserted after text encoding, aimed at removing implicit biases in text input and achieving fairer image generation.
Fair Training with Zero Inputs
Wenjie Pan (Huaqiao University), Huanqiang Zeng (Huaqiao University)
ClassificationRecognitionSegmentationConvolutional Neural NetworkTransformerImage
🎯 What it does: By introducing category-agnostic all-zero inputs and incorporating uniformity loss, the fairness and robustness of classification models are enhanced.
FairGP: A Scalable and Fair Graph Transformer Using Graph Partitioning
Renqiang Luo (Dalian University of Technology), Feng Xia (RMIT University)
Computational EfficiencyGraph Neural NetworkTransformerGraph
🎯 What it does: This paper proposes FairGP, a scalable and fair graph transformer that utilizes graph partitioning to enhance prediction fairness and computational efficiency on large-scale graph data.
Fairness Shields: Safeguarding against Biased Decision Makers
Filip Cano (Graz University of Technology), Kaushik Mallik (IMDEA Software Institute)
OptimizationTabularFinance Related
🎯 What it does: A runtime fairness protection device called the fairness barrier is proposed to minimally intervene with existing black-box decision-makers during sequential decision-making processes, ensuring group fairness metrics are met at limited time windows or periodic time points.
Fairness-Accuracy Trade-Offs: A Causal Perspective
Drago Plecko, Elias Bareinboim (Columbia University)
TabularFinance Related
🎯 What it does: This paper analyzes the trade-off between fairness and accuracy from a causal perspective, and proposes Path-Specific Excess Loss (PSEL) and the corresponding Causal Fairness Utility Ratio (CFUR) to quantify the improvement in fairness and the increase in prediction error under different causal paths.
FairTP: A Prolonged Fairness Framework for Traffic Prediction
Jiangnan Xia (Central South University), Jiannong Cao (Hong Kong Polytechnic University)
Graph Neural NetworkTime Series
🎯 What it does: The FairTP framework is proposed, which studies long-term fairness in traffic prediction and achieves fair predictions at the regional and sensor levels through state recognition and dynamic sampling.
Faithful and Accurate Self-Attention Attribution for Message Passing Neural Networks via the Computation Tree Viewpoint
Yong-Min Shin (Yonsei University), Won-Yong Shin (Yonsei University)
Explainability and InterpretabilityGraph Neural NetworkGraph
🎯 What it does: A method for edge attribution based on the computational tree perspective of self-attention message passing neural networks (Att-GNN) called GATT is proposed;
FakeDiffer: Distributional Disparity Learning on Differentiated Reconstruction for Face Forgery Detection
Bo Wang (Hefei University of Technology), Meng Wang (Hefei University of Technology)
ClassificationRecognitionConvolutional Neural NetworkGenerative Adversarial NetworkImage
🎯 What it does: The FakeDiffer framework is proposed to capture the distribution differences between real and fake images through differential reconstruction learning, thereby enhancing the generalization ability of facial forgery detection.
Falcon: Faster and Parallel Inference of Large Language Models Through Enhanced Semi-Autoregressive Drafting and Custom-Designed Decoding Tree
Xiangxiang Gao (Bestpay AI Lab), Feng Ji (Bestpay AI Lab)
Computational EfficiencyKnowledge DistillationTransformerLarge Language ModelTextBenchmark
🎯 What it does: Proposes the Falcon framework, which improves semi-autoregressive (SAR) inference by combining Coupled Sequential Glancing Distillation with a custom decoding tree to achieve fast, parallel LLM inference.
FAMNet: Frequency-aware Matching Network for Cross-domain Few-shot Medical Image Segmentation
Yuntian Bo (Nanjing University of Science and Technology), Haofeng Zhang (Nanjing University of Science and Technology)
SegmentationDomain AdaptationMeta LearningConvolutional Neural NetworkSupervised Fine-TuningImageBiomedical DataMagnetic Resonance ImagingComputed Tomography
🎯 What it does: A cross-domain few-shot medical image segmentation model, FAMNet, is proposed to address the domain transfer issues caused by different imaging techniques.
FashionTailor: Controllable Clothing Editing for Human Images with Appearance Preserving
Jie Hou (Harbin Institute of Technology), Zhao Zhang (Hefei University of Technology)
Image TranslationGenerationTransformerDiffusion modelImageText
🎯 What it does: A task called 'Clothing Structure Editing (CSE)' is proposed, which allows for the editing of local structures of clothing on human figures based on textual instructions while maintaining the consistency of clothing textures with the human figure. It also achieves control over multiple garments and multiple parts.
Fast and Interpretable Mixed-Integer Linear Program Solving by Learning Model Reduction
Yixuan Li (Southeast University), Wanyuan Wang (Southeast University)
OptimizationExplainability and InterpretabilityComputational EfficiencyTransformerTabular
🎯 What it does: The research achieves fast and interpretable solutions by learning the reducible models of MILP (tight constraint sets and integer variable values);
Fast and Slow Gradient Approximation for Binary Neural Network Optimization
Xinquan Chen (Harbin Institute of Technology), Pengfei Li (Harbin Institute of Technology)
OptimizationConvolutional Neural NetworkRecurrent Neural NetworkImage
🎯 What it does: The Fast and Slow Gradient Generation (FSG) framework is proposed, which approximates the gradient of the non-differentiable quantization function in binary neural network (BNN) training through a dual-branch super network, and introduces Historical Gradient Storage (HGS) and Layer Recognition Embedding (LRE) to enhance gradient quality.
Fast Computing of Dung Semantics in Acyclic Probabilistic Argumentation Frameworks
Stefano Bistarelli (University of Perugia), Carlo Taticchi (University of Perugia)
Computational EfficiencyGraph
🎯 What it does: This paper proposes a method to quickly and accurately calculate the acceptance probabilities of arguments in acyclic (SCG, DAG) abstract argumentation frameworks using Dung semantics within the framework of probability theory (Constellation perspective).
Fast Contiguous Somatic Hypermutations for Single-Objective Optimisation and Multi-Objective Optimisation Via Decomposition
Dogan Corus (Istanbul Bilgi University), Donya Yazdani (British Antarctic Survey)
Optimization
🎯 What it does: This paper proposes an improved continuous cell mutation operator (Fast CHM) and analyzes its runtime in single-objective and multi-objective optimization problems.
Fast Incomplete Multi-view Clustering with Adaptive Similarity Completion and Reconstruction
Deng Xu (Nanjing University), Huaxiong Li (Nanjing University)
OptimizationComputational EfficiencyMultimodality
🎯 What it does: This paper proposes a fast incomplete multi-view clustering method called ASCR, which unifies anchor point learning, adaptive completion and reconstruction of the anchor point-sample similarity graph, and multi-view latent embedding learning, thereby achieving efficient clustering of missing samples.
Fast Multi-Instance Partial-Label Learning
Yin-Fang Yang (Southeast University), Min-Ling Zhang (Southeast University)
ClassificationComputational EfficiencyImageBenchmark
🎯 What it does: A new FASTMIPL framework is proposed for efficient learning from multi-instance partial label data.